Big round of changes to UCD tables, no longer need runtime conversion.

Big round of changes to UCD tables, no longer need runtime conversion.

Previous tables had to be compiled using 9.2, and couldn't use bytevector,
unicode string syntax, or R7RS char/string escapes.  Now they do that which
results in some small space savings and less work at cold load.

Additionally I changed the representation of multiple-code-point values from
strings to vectors.  The old representation depended on the fact that
vector->string didn't do normalization, so those strings can't be constructed at
compile time (they'll be normalized) and anyway it's not good to depend on that
feature of vector->string.